<p>Reliable defect detection in urban sewer networks is critical for infrastructure maintenance, yet sonar inspection in turbid, water-filled pipelines often yields low-contrast, noisy images with blurred boundaries, which reduces detection reliability and efficiency. This study proposes BiGSSM, a sonar defect detector based on a bidirectional gated state-space architecture. The framework integrates bidirectional two-dimensional scanning with adaptive gating to fuse global context and local detail, lightweight depthwise separable convolution blocks to enhance channel-wise representation with limited overhead, and a dynamic focus loss to better handle scale variation, class imbalance, and long-tail distributions. BiGSSM was evaluated on a proprietary Urban Sewer Sonar Defect Dataset and the public Underwater Acoustic Target Detection benchmark. The proposed method achieved mAP50 values of 94.5% and 98.8% on the two datasets, respectively, and consistently outperformed strong baseline detectors in detection accuracy, robustness, and cross-dataset generalization. The results indicate that BiGSSM improves automated sewer sonar defect detection under challenging imaging conditions and can reduce reliance on manual interpretation, supporting scalable inspection of large metropolitan sewer systems.</p>